iommu/core: split mapping to page sizes as supported by the hardware

When mapping a memory region, split it to page sizes as supported
by the iommu hardware. Always prefer bigger pages, when possible,
in order to reduce the TLB pressure.

The logic to do that is now added to the IOMMU core, so neither the iommu
drivers themselves nor users of the IOMMU API have to duplicate it.

This allows a more lenient granularity of mappings; traditionally the
IOMMU API took 'order' (of a page) as a mapping size, and directly let
the low level iommu drivers handle the mapping, but now that the IOMMU
core can split arbitrary memory regions into pages, we can remove this
limitation, so users don't have to split those regions by themselves.

Currently the supported page sizes are advertised once and they then
remain static. That works well for OMAP and MSM but it would probably
not fly well with intel's hardware, where the page size capabilities
seem to have the potential to be different between several DMA
remapping devices.

register_iommu() currently sets a default pgsize behavior, so we can convert
the IOMMU drivers in subsequent patches. After all the drivers
are converted, the temporary default settings will be removed.

Mainline users of the IOMMU API (kvm and omap-iovmm) are adopted
to deal with bytes instead of page order.

Many thanks to Joerg Roedel <Joerg.Roedel@amd.com> for significant review!
Signed-off-by: Ohad Ben-Cohen <ohad@wizery.com>
Cc: David Brown <davidb@codeaurora.org>
Cc: David Woodhouse <dwmw2@infradead.org>
Cc: Joerg Roedel <Joerg.Roedel@amd.com>
Cc: Stepan Moskovchenko <stepanm@codeaurora.org>
Cc: KyongHo Cho <pullip.cho@samsung.com>
Cc: Hiroshi DOYU <hdoyu@nvidia.com>
Cc: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Cc: kvm@vger.kernel.org
Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>

drivers/iommu/iommu.c

@@ -16,6 +16,8 @@
  * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307 USA
  */
 
+#define pr_fmt(fmt)    "%s: " fmt, __func__
+
 #include <linux/device.h>
 #include <linux/kernel.h>
 #include <linux/bug.h>
@@ -47,6 +49,16 @@ int bus_set_iommu(struct bus_type *bus, struct iommu_ops *ops)
 	if (bus->iommu_ops != NULL)
 		return -EBUSY;
 
+	/*
+	 * Set the default pgsize values, which retain the existing
+	 * IOMMU API behavior: drivers will be called to map
+	 * regions that are sized/aligned to order of 4KiB pages.
+	 *
+	 * This will be removed once all drivers are migrated.
+	 */
+	if (!ops->pgsize_bitmap)
+		ops->pgsize_bitmap = ~0xFFFUL;
+
 	bus->iommu_ops = ops;
 
 	/* Do IOMMU specific setup for this bus-type */
@@ -157,34 +169,125 @@ int iommu_domain_has_cap(struct iommu_domain *domain,
 EXPORT_SYMBOL_GPL(iommu_domain_has_cap);
 
 int iommu_map(struct iommu_domain *domain, unsigned long iova,
-	      phys_addr_t paddr, int gfp_order, int prot)
+	      phys_addr_t paddr, size_t size, int prot)
 {
-	size_t size;
+	unsigned long orig_iova = iova;
+	unsigned int min_pagesz;
+	size_t orig_size = size;
+	int ret = 0;
 
 	if (unlikely(domain->ops->map == NULL))
 		return -ENODEV;
 
-	size         = PAGE_SIZE << gfp_order;
+	/* find out the minimum page size supported */
+	min_pagesz = 1 << __ffs(domain->ops->pgsize_bitmap);
+
+	/*
+	 * both the virtual address and the physical one, as well as
+	 * the size of the mapping, must be aligned (at least) to the
+	 * size of the smallest page supported by the hardware
+	 */
+	if (!IS_ALIGNED(iova | paddr | size, min_pagesz)) {
+		pr_err("unaligned: iova 0x%lx pa 0x%lx size 0x%lx min_pagesz "
+			"0x%x\n", iova, (unsigned long)paddr,
+			(unsigned long)size, min_pagesz);
+		return -EINVAL;
+	}
+
+	pr_debug("map: iova 0x%lx pa 0x%lx size 0x%lx\n", iova,
+				(unsigned long)paddr, (unsigned long)size);
+
+	while (size) {
+		unsigned long pgsize, addr_merge = iova | paddr;
+		unsigned int pgsize_idx;
+
+		/* Max page size that still fits into 'size' */
+		pgsize_idx = __fls(size);
+
+		/* need to consider alignment requirements ? */
+		if (likely(addr_merge)) {
+			/* Max page size allowed by both iova and paddr */
+			unsigned int align_pgsize_idx = __ffs(addr_merge);
+
+			pgsize_idx = min(pgsize_idx, align_pgsize_idx);
+		}
+
+		/* build a mask of acceptable page sizes */
+		pgsize = (1UL << (pgsize_idx + 1)) - 1;
 
-	BUG_ON(!IS_ALIGNED(iova | paddr, size));
+		/* throw away page sizes not supported by the hardware */
+		pgsize &= domain->ops->pgsize_bitmap;
 
-	return domain->ops->map(domain, iova, paddr, size, prot);
+		/* make sure we're still sane */
+		BUG_ON(!pgsize);
+
+		/* pick the biggest page */
+		pgsize_idx = __fls(pgsize);
+		pgsize = 1UL << pgsize_idx;
+
+		pr_debug("mapping: iova 0x%lx pa 0x%lx pgsize %lu\n", iova,
+					(unsigned long)paddr, pgsize);
+
+		ret = domain->ops->map(domain, iova, paddr, pgsize, prot);
+		if (ret)
+			break;
+
+		iova += pgsize;
+		paddr += pgsize;
+		size -= pgsize;
+	}
+
+	/* unroll mapping in case something went wrong */
+	if (ret)
+		iommu_unmap(domain, orig_iova, orig_size - size);
+
+	return ret;
 }
 EXPORT_SYMBOL_GPL(iommu_map);
 
-int iommu_unmap(struct iommu_domain *domain, unsigned long iova, int gfp_order)
+size_t iommu_unmap(struct iommu_domain *domain, unsigned long iova, size_t size)
 {
-	size_t size, unmapped;
+	size_t unmapped_page, unmapped = 0;
+	unsigned int min_pagesz;
 
 	if (unlikely(domain->ops->unmap == NULL))
 		return -ENODEV;
 
-	size         = PAGE_SIZE << gfp_order;
+	/* find out the minimum page size supported */
+	min_pagesz = 1 << __ffs(domain->ops->pgsize_bitmap);
+
+	/*
+	 * The virtual address, as well as the size of the mapping, must be
+	 * aligned (at least) to the size of the smallest page supported
+	 * by the hardware
+	 */
+	if (!IS_ALIGNED(iova | size, min_pagesz)) {
+		pr_err("unaligned: iova 0x%lx size 0x%lx min_pagesz 0x%x\n",
+					iova, (unsigned long)size, min_pagesz);
+		return -EINVAL;
+	}
+
+	pr_debug("unmap this: iova 0x%lx size 0x%lx\n", iova,
+							(unsigned long)size);
+
+	/*
+	 * Keep iterating until we either unmap 'size' bytes (or more)
+	 * or we hit an area that isn't mapped.
+	 */
+	while (unmapped < size) {
+		size_t left = size - unmapped;
+
+		unmapped_page = domain->ops->unmap(domain, iova, left);
+		if (!unmapped_page)
+			break;
 
-	BUG_ON(!IS_ALIGNED(iova, size));
+		pr_debug("unmapped: iova 0x%lx size %lx\n", iova,
+					(unsigned long)unmapped_page);
 
-	unmapped = domain->ops->unmap(domain, iova, size);
+		iova += unmapped_page;
+		unmapped += unmapped_page;
+	}
 
-	return get_order(unmapped);
+	return unmapped;
 }
 EXPORT_SYMBOL_GPL(iommu_unmap);

drivers/iommu/omap-iovmm.c

@@ -409,7 +409,6 @@ static int map_iovm_area(struct iommu_domain *domain, struct iovm_struct *new,
 	unsigned int i, j;
 	struct scatterlist *sg;
 	u32 da = new->da_start;
-	int order;
 
 	if (!domain || !sgt)
 		return -EINVAL;
@@ -428,12 +427,10 @@ static int map_iovm_area(struct iommu_domain *domain, struct iovm_struct *new,
 		if (bytes_to_iopgsz(bytes) < 0)
 			goto err_out;
 
-		order = get_order(bytes);
-
 		pr_debug("%s: [%d] %08x %08x(%x)\n", __func__,
 			 i, da, pa, bytes);
 
-		err = iommu_map(domain, da, pa, order, flags);
+		err = iommu_map(domain, da, pa, bytes, flags);
 		if (err)
 			goto err_out;
 
@@ -448,10 +445,9 @@ static int map_iovm_area(struct iommu_domain *domain, struct iovm_struct *new,
 		size_t bytes;
 
 		bytes = sg->length + sg->offset;
-		order = get_order(bytes);
 
 		/* ignore failures.. we're already handling one */
-		iommu_unmap(domain, da, order);
+		iommu_unmap(domain, da, bytes);
 
 		da += bytes;
 	}
@@ -466,7 +462,8 @@ static void unmap_iovm_area(struct iommu_domain *domain, struct omap_iommu *obj,
 	size_t total = area->da_end - area->da_start;
 	const struct sg_table *sgt = area->sgt;
 	struct scatterlist *sg;
-	int i, err;
+	int i;
+	size_t unmapped;
 
 	BUG_ON(!sgtable_ok(sgt));
 	BUG_ON((!total) || !IS_ALIGNED(total, PAGE_SIZE));
@@ -474,13 +471,11 @@ static void unmap_iovm_area(struct iommu_domain *domain, struct omap_iommu *obj,
 	start = area->da_start;
 	for_each_sg(sgt->sgl, sg, sgt->nents, i) {
 		size_t bytes;
-		int order;
 
 		bytes = sg->length + sg->offset;
-		order = get_order(bytes);
 
-		err = iommu_unmap(domain, start, order);
-		if (err < 0)
+		unmapped = iommu_unmap(domain, start, bytes);
+		if (unmapped < bytes)
 			break;
 
 		dev_dbg(obj->dev, "%s: unmap %08x(%x) %08x\n",

include/linux/iommu.h

@@ -48,6 +48,19 @@ struct iommu_domain {
 
 #ifdef CONFIG_IOMMU_API
 
+/**
+ * struct iommu_ops - iommu ops and capabilities
+ * @domain_init: init iommu domain
+ * @domain_destroy: destroy iommu domain
+ * @attach_dev: attach device to an iommu domain
+ * @detach_dev: detach device from an iommu domain
+ * @map: map a physically contiguous memory region to an iommu domain
+ * @unmap: unmap a physically contiguous memory region from an iommu domain
+ * @iova_to_phys: translate iova to physical address
+ * @domain_has_cap: domain capabilities query
+ * @commit: commit iommu domain
+ * @pgsize_bitmap: bitmap of supported page sizes
+ */
 struct iommu_ops {
 	int (*domain_init)(struct iommu_domain *domain);
 	void (*domain_destroy)(struct iommu_domain *domain);
@@ -61,6 +74,7 @@ struct iommu_ops {
 				    unsigned long iova);
 	int (*domain_has_cap)(struct iommu_domain *domain,
 			      unsigned long cap);
+	unsigned long pgsize_bitmap;
 };
 
 extern int bus_set_iommu(struct bus_type *bus, struct iommu_ops *ops);
@@ -72,9 +86,9 @@ extern int iommu_attach_device(struct iommu_domain *domain,
 extern void iommu_detach_device(struct iommu_domain *domain,
 				struct device *dev);
 extern int iommu_map(struct iommu_domain *domain, unsigned long iova,
-		     phys_addr_t paddr, int gfp_order, int prot);
-extern int iommu_unmap(struct iommu_domain *domain, unsigned long iova,
-		       int gfp_order);
+		     phys_addr_t paddr, size_t size, int prot);
+extern size_t iommu_unmap(struct iommu_domain *domain, unsigned long iova,
+		       size_t size);
 extern phys_addr_t iommu_iova_to_phys(struct iommu_domain *domain,
 				      unsigned long iova);
 extern int iommu_domain_has_cap(struct iommu_domain *domain,

virt/kvm/iommu.c

@@ -113,7 +113,7 @@ int kvm_iommu_map_pages(struct kvm *kvm, struct kvm_memory_slot *slot)
 
 		/* Map into IO address space */
 		r = iommu_map(domain, gfn_to_gpa(gfn), pfn_to_hpa(pfn),
-			      get_order(page_size), flags);
+			      page_size, flags);
 		if (r) {
 			printk(KERN_ERR "kvm_iommu_map_address:"
 			       "iommu failed to map pfn=%llx\n", pfn);
@@ -292,15 +292,15 @@ static void kvm_iommu_put_pages(struct kvm *kvm,
 
 	while (gfn < end_gfn) {
 		unsigned long unmap_pages;
-		int order;
+		size_t size;
 
 		/* Get physical address */
 		phys = iommu_iova_to_phys(domain, gfn_to_gpa(gfn));
 		pfn  = phys >> PAGE_SHIFT;
 
 		/* Unmap address from IO address space */
-		order       = iommu_unmap(domain, gfn_to_gpa(gfn), 0);
-		unmap_pages = 1ULL << order;
+		size       = iommu_unmap(domain, gfn_to_gpa(gfn), PAGE_SIZE);
+		unmap_pages = 1ULL << get_order(size);
 
 		/* Unpin all pages we just unmapped to not leak any memory */
 		kvm_unpin_pages(kvm, pfn, unmap_pages);